The Magnetic Resonance (MR) Scanner is typically connected to a computer or workstation, which is used to obtain process and store image data which can be processed and displayed as images. The image data is captured as raw data (k-space data or magnetic resonance data) that contains artifacts inherent to the scanner.
During MR reconstruction (when the patient is in the scan room), the raw image data (k-space data) undergoes a set of transformations to convert the images in k-space to image space (spatial domain) before they are displayed on a monitor. The medical professional then reviews the images for a patient or saves the images and recalls them later for more diagnosis.
Image processing algorithms are applied to the raw image data to convert them from k-space to image space so that it can be viewed, analyzed and diagnosed by the medical professional. There are many image processing algorithms that add diagnostic value to these images. Image processing is a research field in itself, with many algorithms of high diagnostic value being continually developed. New image processing algorithms are often developed and integrated into the reconstruction chain.
The image processing steps that needs to be applied depends on the scan type (like Spin Echo scan, Gradient Echo scan, SENSE scan, Dixon scan etc). Currently as the scan time of MR scanner is considerably high there is continuous research going on to come up with new scan type which would reduce the scan time. Often whenever a new scan type is introduced it would warrant for a new image processing algorithm in reconstruction chain.
For fast execution of image reconstruction the computer resources like the CPU (cores), memory will have to be used judiciously and efficiently.
Currently MR image reconstruction is developed as a monolithic and closed package in a programming language like C++ without the clear separation of mathematical numerical processing and software engineering disciplines. This application is compiled for distribution to customers who purchase, lease or own the MR Scanner. Integration of new algorithms requires recompiling the application software. Also the number of threads to be used by the application is hardcoded in the application. The off-the-shelf libraries used are hardcoded in the code and any change in the library needs code modification and recompilation of the application software.
United States patent application US 2005/0018929 A1 discloses a method for dynamically editing and enhancing image-processing chains in medical imaging equipment.